Yes — with 8.0 GB to spare

DeepSeek-R1-Distill-Qwen 14B at Q4_K_M fits your RTX 4000 Ada entirely on the GPU at 8K context, at an estimated 26 tokens per second. Past 50K the KV cache pushes it over — quantise the cache to q8_0, or step down a quantisation, to go longer.

Fully on GPU 8K context Q4_K_M · 8.3 GB MIT Released Jan 2025

The 2025 reasoning-per-gigabyte pick for a 12 GB card. Qwen3.5 9B in thinking mode has since overtaken it.

What hardware do I need for DeepSeek-R1-Distill-Qwen 14B? →

The VRAM budget

weights 8.3 GB

Weights 8.3 GB KV cache @ 8K 1.50 GB Runtime overhead 0.6 GB Free 8.0 GB of 18.4 GB

Quantisation ladder

Quant	Weights	Total @ 8K	Max context	Tok/s	Quality	Fit
F16	27.6 GB	29.7 GB	—	~2.6	Reference	11.3 GB over
Q8_0	14.6 GB	16.7 GB	16K	15	−0.1% ppl	Fits
Q6_K	11.3 GB	13.4 GB	34K	19	−0.4% ppl	Long context
Q5_K_M	9.8 GB	11.9 GB	42K	22	−0.8% ppl	Long context
Q4_K_M	8.3 GB	10.4 GB	50K	26	−1.9% ppl	Recommended
Q3_K_M	6.7 GB	8.8 GB	59K	32	−5.4% ppl	Long context
Q2_K	5.8 GB	7.9 GB	64K	38	−15% ppl	Long context

Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it.

How to run it

Ollama llama.cpp LM Studio vLLM

terminal

$ ollama pull deepseek-r1:14b
$ OLLAMA_CONTEXT_LENGTH=8192 \
    ollama run deepseek-r1:14b

The default. One binary, a model registry, an OpenAI-compatible port. More on Ollama.

01Download is 8.3 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.

02Close anything else holding VRAM. A browser with hardware acceleration can sit on 1–2 GB.

03There is room to go to 50K context on this card.

See all models for this rig Compare with another model